Research On Trust Computing Based On ELM

With the large scale of soial network members,there is a sparse direct trust relationship among the users.Therefore,better sevices and higher security of user communication can be acquired through the indirect trust of the online social network(OSN).At the same time,in applications such as auxiliary decision-making,recommendation system and privacy protection,trust also counts.The most common method to study the process of trust propagation and aggregation in OSN is what scholars called deduction.However,it is based on congective experience and hard to be applied to various social networks.Some scholars use neural networks based on training to study the calculation of trust,but the certainty of neural network as well as the uncertainty of the topology among users make it difficult to simulate the process of trust propagation and aggregation among users through the use of feedforward neural network in OSN.Directed by the theory of extreme learning machine(ELM)and multi-layer extreme learning machine(ML-ELM),this thesis proposes two neural network models based on training,ELM-Walk Net and ML-ELM-Walk Net,which can be used to model the calculation process of two-step trust among users.Fistly,the two models put all the trust values on the two-step trust path between two users into a vector as the input data of the neural network.Secondly,simulating the two-step trust process through the outputs accessed by a single hidden layer feedforward neural network and a multi-hidden layer feedforward neural network respectively.Thirdly,using the two models to calculate the two-step trust among all users in OSN,and all the calculated trust values are added to OSN.Lastly,ELM-NeuralWalk and ML-ELM-NeuralWalk cyclically call ELM-Walk Net and ML-ELM-Walk Net respectively to complete the multi-step trust calculation among all users in OSN,which solves the problem of topology inconsistency among users in OSN.In addition,ML-ELM-NeuralWalk finds training data and prediction data through Boolean matrix operation,which effectively improves the calculation speed of training set and prediction set.The experiment turns out that,compared with the existing trust computing solutions,the two neural network models put forward in this thesis simulate the two-step trust computing process among users more accurately.Moreover,more efficient multi-step trust computing results can be provided with the aid of ELM-NeuralWalk and ML-ELM-NeuralWalk algorithms.